Abstract: In view of the problem that the load state of the ball mill cannot be accurately characterized by a single-signal characteristic parameter, statistical analysis and AR model are used to help extract the time domain characteristic parameter and frequency domain characteristic parameter of the ball mill's vibration signal. Then, the D-S evidence theory is also adopted to analyze the two parameters, and the basic probability distribution function of each characteristic parameter is calculated. Finally, the basic probability distribution function value is taken as the characteristic value of the ball mill in load state, and the characteristic vector model of multi-vibration characteristic parameter is established. The results show that this model can well characterize the load state of the ball mill and provide data supporting for the prediction of the load of the ball mill.